Landing: 465ee014cb0c
Project / Subsystem
gcc / gcc/middle-end
Date
2026-08-13
Author
Richard Biener
Commit
465ee014cb0cf9be62f0fd3de3821c755d37f58d
Source
github
Perf win
Yes
Breaking
No
All attributes
- project
- gcc
- subsystem
- gcc/middle-end
- patch_id
- —
- commit_hash
- 465ee014cb0cf9be62f0fd3de3821c755d37f58d
- source_type
- github
- headline
- Improve vector conversion folding by handling NOP and sign conversions correctly.
- tldr
- Vector conversion optimizations now correctly handle no-operation and sign-preserving conversions.
- author
- Richard Biener
- outcome
- committed
- performance_win
- true
- breaking_change
- false
- series_id
- —
- series_parts
- []
- tags
-
- • optimization
- • vectorization
- • middle-end
- • type conversion
- discussion_id_link
- —
- bugzilla_pr
- —
- date
- 2026-08-13T00:00:00.000Z
This commit refines vector conversion optimizations by addressing issues in supportable_convert_operation and vectorizable_conversion. It ensures that no-operation (NOP) conversions and sign-preserving conversions are correctly handled, preventing test suite failures. The changes allow NOP_EXPR and VIEW_CONVERT_EXPR for same-mode types, and modify vectorizable_conversion to prefer VIEW_CONVERT_EXPR for vector types, ensuring that these conversions are recognized and optimized appropriately without introducing incorrect behavior.